.1 and .2 were on the same Centos 4 system, built with gcc 3. I've not yet built a local copy of .0, that was checked against the system perl on a SunOS box. I'll try to look into that soonish, and hope it's not bitten by the same compilation issues as 5.6.1 under gcc 3; one must munge the makefile's to remove <built-in> and <command line>, etc.
My original guess was something to do with locales, until I noticed that all of the CPANTS failures were 5.6.2 boxen.
In Bob We Trust, All Others Bring Data.